Evaluation of pre-processing tools and provenance in RNA-Seq studies of breast cancer

Gene expression profiling is instrumental in diseases such as breast cancer to tailor treatment approaches based on different subtypes of tumors. Reproducibility in genomic studies is essential because the results are being used in drug design, diagnostics, and patient care. When results are not reproducible in genomic studies, the reliability and validity of these applications are lost. Achieving reproducibility is by ensuring that provenance is written in every manuscript. Reproducibility is the ability to come up with the same results with the same data at another time, while provenance is the documentation of how the data was manipulated to ensure that the results would always be the same. For RNA-Seq papers, provenance includes having sufficient information on the pipeline employed. Aside from this, there is a myriad of modifications in tools to manipulate the data in RNA-Seq such as the steps of preprocessing, alignment, the genomic annotation reference, and the quantification. In this study, the provenance of breast cancer RNA-Seq papers was examined, and the results of three preprocessing and two alignment tools were compared. A total of 106 breast cancer RNA-Seq papers were examined for provenance criteria. In the 65 primary data papers out of the 106 breast cancer RNA-Seq papers, preprocessing tools (23%) were the least reported followed by dataset (65%), reference for genomic annotation (68%), quantification tool (75%), alignment type (82%), and alignment tools (88%). It was found that the number of reads used for downstream analysis and whether it is pair-ended or single-ended is mentioned. However, the version of the tool, the trimming option, the number of discards, and whether the samples are biological replicates or pooled data are often unreported. From the examined breast cancer RNA-Seq studies most common pre-processing tools involved in trimming reads were Trimmomatic, Cutadapt, and Trim Galore. Comparing the three tools, Cutadapt and Trim Galore gives a 1~2% improvement in alignment scores because of the algorithm they share. Upon close examination of genes BRCA1 and BRCA2, HISAT2 and RNA Star have a discrepancy alignment rate of 19% where reads in one tool are not aligned in the other. Cutadapt and HISAT2 were more efficient in the use of computer resources based on processing times in Galaxy. Unreported tools are a concern as different results can still be generated despite using the same dataset and reference genome. This paper underscores the importance of ensuring that published papers have proper provenance given these findings.